coach/benchmarks/dueling_ddqn_with_per/README.md

Dueling DDQN with Prioritized Experience Replay

Each experiment uses 3 seeds and is trained for 10k environment steps. The parameters used for Dueling DDQN with PER are the same parameters as described in the following paper.

Breakout Dueling DDQN with PER - single worker

python3 coach.py -p Atari_Dueling_DDQN_with_PER_OpenAI -lvl breakout

Pong Dueling DDQN with PER - single worker

python3 coach.py -p Atari_Dueling_DDQN_with_PER_OpenAI -lvl pong

Space Invaders Dueling DDQN with PER - single worker

python3 coach.py -p Atari_Dueling_DDQN_with_PER_OpenAI -lvl space_invaders